Numerical Design of Experiments for Repeating Low-Pressure Turbine Stages Part I: Computational Opportunities and Methodology

Abstract The trend towards more compact and efficient low-pressure turbine (LPT) designs can substantially benefit from advanced numerical predictive tools. The complex transitional and turbulent nature of unsteady flows seen in LPTs often demands highorder methods such as Large Eddy Simulations (LES) for accurate predictions in turbine efficiency and loss generation. Integrating high-fidelity simulations into design cycles, predominately driven by rapid Unsteady Reynolds-Averaged Navier-Stokes (URANS) calculations, requires cutting-edge numerical tools able to leverage modern high-performance computing architectures. In Part I of this paper, a multi-fidelity simulation framework is presented, which maximizes the computational output of the latest highperformance architectures by fully occupying the hardware with concurrent LES and URANS simulations. The continuous rise in computational power of supercomputing facilities has primarily been driven by advances in Graphics Processing Units (GPUs). GPUs offer substantial parallelism and enable the rapid processing of large datasets which is ideally suited for performing LESs requiring extensive grid counts for adequate resolution. In contrast, modern Central Processing Units (CPUs) typically have between 32 and 64 cores, offering lower performance than GPUs, and are better suited for less compute-intensive tasks, such as Reynolds-Averaged Navier-Stokes (RANS) calculations. While previous studies have focused on numerical methods optimized for either CPU or GPU architectures, this paper introduces a novel multi-fidelity approach using both the CPUs and GPUs on the same computational node concurrently, increasing the utilization of modern supercomputers. In this framework, high Reynolds number LESs are executed on the GPUs, while the otherwise idling CPU cores are used for multiple URANS calculations, or an additional low Reynolds number LES case. As a result, this approach maximizes the efficiency and computational output of the entire node, and provides high-fidelity and low-fidelity results at the same time. Ultimately, this paper combines the low-cost trend predictions of URANS with the accuracy of LES to create a multi-fidelity dataset spanning the entire Reynolds number regime of interest via an established multi-fidelity reconstruction method. The multi-fidelity reconstructions are validated against LES data not considered in the original dataset, and the solutions prove superior to URANS at all operating conditions, minimizing the number of costly LESs required for a highly accurate and fine-grained parametric sweep suitable for industrial design cycles.